OT: Silicon oxide as a switch

Intel thinks on-chip optical interconnects have potential. (Kobrinsky, et al, "On-Chip Optical Interconnects", Intel Technology Journal, v. 8, n. 2, May 10, 2004) Maybe not so much for clock, where they don't see significant advantages compared to metal interconnect, but definitely for signaling. I could be wrong, but I think Intel has a clue about semiconductor technology. Other people (Ackland, et al, "A Comparison of Electrical and Optical Clock Networks in Nanometer Technologies", IEEE 2005 Custom Integrated Circuit Conference) point out some flaws in the Kobrinsky analysis, and are more positive about the possibilities of using on-chip optical interconnects for clock.

It's not necessary to use IR wavelengths for an on-chip system. I've done work on on-chip optical interconnects using 654nm red laser diodes (Thangaraj C., et al, "Fully CMOS-Compatible on-Chip Optical Clock Distribution & Recovery", IEEE Trans. VLSI Systems, accepted for publication; Pownall, R., et al, "CMOS optoelectronic components for clock distribution", Microelectronic Engineering, v87, n10, pp.

1838-1845 (Oct. 2010) and Pownall, R., et al, "AC Performance of Polysilicon Leaky-Mode MSM Photodetectors", accepted for publication in IEEE J. Lightwave Tech.) We've achieved success optically clocking actual electronic circuits. They didn't run as fast as we wanted, but that was due to underestimation of circuit parasitics when doing the circuit design, not the performance of the optoelectronic parts of the system. The performance of the waveguides and detectors was up in the few hundred MHz range. Yes, that's a factor of 10 slower than needed, but we were using a plain-vanilla CMOS process with as few changes as possible. We're pretty sure we know what we need to do in order to get faster performance - people have demonstrated GHz performance out of detectors similar to ours - but we've run out of money.

Yes, it's possible that we and everyone else looking into on-chip optical interconnects are heading down a dead end. However, IMHO it's way too early to come to that conclusion, especially since we've already done stuff that you seem to feel can't be done.

It is not so much that i think it can't be done, but that were far away from having anything more than a niche technology.